I am a paying subscriber for ChatGPT, but I cannot access support. The chat bubble won’t let me type, and the Feedback Form keeps redirecting me to the same broken system. I need urgent help resolving issues with my account and accessing customer service. Please advise.
I am a paying subscriber, and I cannot access support. The chat bubble won’t allow me to type, and the Feedback Form keeps redirecting me to the same system with no options addressing my issue.
I need assistance with repeatedly getting incorrect output from Chatgpt, despite my prompts being very exact. I get continuous typos from the output it just gave me when put into flyer form. I also get locked out and have to open a new tab every single time and I cleared my cache and cookies and have tried from various computers. This is all unacceptable and very timing consuming. I’m wasting so much time. ].
Please help me escalate this to a support agent. Thank you!"
Typically when there’s some sort of discrepancy between the output and expected output it’s from a misunderstanding on how to prompt an LLM, and it’s capabilities.
There’s numerous people here, including myself that are happy to help if you can provide some examples. Otherwise
This is not something we’re capable of doing here. Like everyone else you’ll just have to wait for help to get back. There’s no line skipping.
Ronald, thank you very much for your quick response. Actually, Chatgpt gave me instructions of what to do to report my issues. It was them who said to have my problem escalated to a support agent.
I welcome you helping me to get better output. What happens repeatedly is I give a prompt to make a flyer. For example, I’ll say "give me a graphic for a flyer and include the following information: I am doing a taste testing at Sahadi’s on December 1st, and I’d like you to write me a headline, subheadline, and the body for the flyer saying MAMBA hot sauce will be doing this taste testing. Please make it enticing and put it on a background that is holiday oriented. The graphics are great, but every single time, if they give me the text, and then I ask for the graphic, it never matches what they original wrote. And the typos! OMG! They are endless, so even say specifically : use the text exactly how I write it, and make sure there are no typos. It doesn’t matter. If you could provide me with prompts on how to do this better, I welcome that. Thank you so much!
Ah, ChatGPT and LLMs in general are not good at metadata. As in, if it’s information regarding themselves or the company that made them it will most likely hallucinate information.
Oh yes, typos are a HUGE issue with the image generation tool. This is a well-known issue that’s inherent in diffusion models.
Unfortunately your best bet is to have it generate the same content numerous times and hope that you can apply some photoshop magic to piece it back into the land of coherency.
Ronald, thank you for your quick response, once again.
Please excuse me here, as I am new to this whole AI world, but what is LLM? And I’m not sure what metadata is! The term “hallucinate” is so interesting. Not sure what it means, though.
As far as the typos go, I actually was arguing with Chatgpt, like a crazy person So, this is a known problem in “diffusion” models. What is that? All AI? I heard Midjourney has the same issues.
I have spent over an hour working on the same flyer, with no success. Is Chatgpt fixing this? This is a paid service, so I didn’t anticipate this.
Thanks so much for reaching back out to me. I really appreciate it.
An LLM (Large Language Model) is a sub-category of AI. It’s the category that defines models like GPT.
In terms of metadata: I used it loosely. What would be better for me to say that asking a model about itself, or it’s own company and actions can lead to hallucinations, which is the industry description of a model that “makes up” information.
We’ve all been there Typically once the arguing starts it’s better to just delete the conversation and start fresh. A personal opinion but each time a message is sent and received you are introducing bias to the model and straying further away from it’s training data.
Yes, most popular models that generate images are done through a technique called “diffusion”. Here’s a wonderful video that goes into details about it. If I remember correctly it gets quite technical but the visual transitioning really drives home how it works and hopefully will give some insights into why forming coherent letters is so difficult.
Believe me when I say it was A LOT worse in the past
One tool that springs to mind that may be better is Microsoft Designer. It combines ChatGPT with photoshop-like elements that really help bridge the gap between the two.
So, this is a known problem in “diffusion” models. What is that? All AI? I heard Midjourney has the same issues.
Typically once the arguing starts it’s better to just delete the conversation and start fresh. A personal opinion but each time a message is sent and received you are introducing bias to the model and straying further away from it’s training data.
